silvarion.llm

Ansible collection for deploying and managing local LLM infrastructure.

---

Roles

Role	Description
silvarion.llm.ollama	Installs and manages the Ollama LLM model server
silvarion.llm.openclaw	Installs the OpenClaw personal AI assistant CLI via npm
silvarion.llm.opencode	Installs the OpenCode AI coding assistant CLI
silvarion.llm.open_webui	Installs and manages Open WebUI (web frontend for LLMs)
silvarion.llm.vllm	Installs and manages the vLLM OpenAI-compatible inference server

---

Supported Platforms

Distribution	Versions
Ubuntu	22.04 (jammy), 24.04 (noble)
Debian	12 (bookworm)
Rocky Linux / AlmaLinux / RHEL	9

---

Requirements

• Ansible >= 2.14
• gather_facts: true on all plays
• become: true on all plays

---

Installation

ansible-galaxy collection install silvarion.llm

---

Usage

Deploy full local LLM stack

---
- name: Deploy local LLM stack
  hosts: llm_servers
  gather_facts: true
  become: true
  roles:
    - role: silvarion.llm.ollama
    - role: silvarion.llm.open_webui
      vars:
        openweb_ui_config_ollama_url: "http://localhost:11434"

Install Ollama and pull a model

---
- name: Install Ollama with llama3 model
  hosts: gpu_servers
  gather_facts: true
  become: true
  roles:
    - role: silvarion.llm.ollama
      vars:
        ollama_models_pull:
          - llama3
          - mistral

Install OpenCode CLI

---
- name: Install OpenCode
  hosts: developer_workstations
  gather_facts: true
  become: true
  roles:
    - role: silvarion.llm.opencode

Install OpenClaw AI assistant

---
- name: Install OpenClaw
  hosts: developer_workstations
  gather_facts: true
  become: true
  roles:
    - role: silvarion.llm.openclaw
      vars:
        openclaw_config_model: "ollama/llama3"
        openclaw_config_users:
          - name: alice
            home: /home/alice

Deploy vLLM inference server

---
- name: Deploy vLLM
  hosts: gpu_servers
  gather_facts: true
  become: true
  roles:
    - role: silvarion.llm.vllm
      vars:
        vllm_config_model: "mistralai/Mistral-7B-Instruct-v0.3"
        vllm_huggingface_token: "{{ vault_hf_token }}"
        vllm_config_tensor_parallel_size: 1
        vllm_config_dtype: "bfloat16"

---

License

MIT

Author

Jesus Alejandro Sanchez Davila — jsanchez.consultant@gmail.com